7, 18 Feb 2002, L.A. & DLD.
Topics include: Elementary information theory (including noiseless coding and compression); introduction to inductive inference and prediction; data modelling and data mining; introduction to Minimum Message Length (MML) inference; clustering, mixture modelling and unsupervised classification; supervised classification and decision trees. Example applications may be described.
foundn/models/applics [[practical]] ------------- ------------- 1: prob' (codes), info', terminology, 2: estimators, entropy, KL-dist, prob' pred'n 3: (Snob out of synch') derive binomial ?3-ways? + Fisher, 4: give multinomial maxL MML minEKL, [[do unsup' class'n, mix' model & use Snob]] 5: normal, give Fisher, maxL, MML 6: [Unsupervised] mixture modelling + Snob 7: other prob' distributions 8: [Supervised] decision trees 1 [[do sup' class'n and use a dec' tree]] 9: decision trees 2 10: dist'ns and codes for integers 11: KL defn on a mult, on a normal 12: codes++, Kraft, Huffman, Shannon-Fano; application advert' (for MML:-)
CSE454 is a prereq'
Topics include: Foundations of inductive inference; Minimum Message Length (MML) inference; Fisher information; MML of specific models such as decision graphs, hidden Markov models, linear and polynomial regression, causal models. Data mining. Applications to be considered may include: text and image analysis, models of protein folding, bushfire prediction, DNA alignment and the human genome project, authorship identification for texts, etc.
foundn/models/applics [[practical] ------------- ------------ 1: Kolmo', Solom',..., Fisher (more) derive normal, mention Poisson F' 2: decision graphs 3: time-series, segmentation, trends etc. 4: sequences, H-Markov-Ms, W+Georgeff [[time series, seg'n &/or HMM &/or auto reg']] 5: DNA pattern discovery 6: alignment, evolutionary trees 7: protein secondary structure pred'n 8: lin'-regression, polynomials and/or polygons [[regression, polynomials, polygons, A.N.N.]] 9: mixtures + 1st order Markov 10: causal and/or neural models 11, 12:research applics such as... Lempel-Ziv, Wally improv', approx' repeats images, noisy/dirty pics, Markov fields